Not much information about the hardware randomness source.

Information on how to generate digital TRNG from a reverse-biased PN junction has been all over the internet for a couple of decades. My 2009 design, which simply slotted into an Arduino board, furnished 200kB/s (1.6Mbit/s) of data that passed all the statistical tests I threw at it at the time (mostly diehard(er)).

To do all this in an 8-bit microcontroller running at 16MHz required coding in assembly and conditioning the raw noise with a noncryptographic hash. https://emergent.unpythonic.net/01257868826

Nowadays with an embedded microcontroller like a SAM D51, even if you want to ignore the built in TRNG that can apparently generate over 40Mbit/s (32 bits per 84 cycles, CPU runs at 120MHz) and meets NIST / FIPS requirements, you could do much better by sampling random data more frequently and using a cryptographic conditioning function.